On Wed, 16 Oct 2024 at 03:40, Robert Haas <robertmh...@gmail.com> wrote: > > On Sat, Oct 12, 2024 at 5:56 AM Shayon Mukherjee <shay...@gmail.com> wrote: > > Thank you for sharing this Robert. I like the idea behind "allowing > > extensions to control planner behavior” overall and I think it does help > > towards a powerful extension ecosystem too. I wonder if there is a reality > > where we can achieve both the outcomes here > > > > - Support disabling of indexes [1] through ALTER command > > - While also building on "allowing extensions to control planner behavior” > > for the reasons above? > > > > [1] > > https://www.postgresql.org/message-id/ABD42A12-4DCF-4EE4-B903-4C657903CECE%40gmail.com > > Yes, I think we can do both things.
I think so too. I imagine there'd be cases where even hints global to all queries running on the server wouldn't result in the index being completely disabled. For example, a physical replica might not be privy to the hints defined on the primary and it might just be the queries running on the physical replica that are getting the most use out of the given index. Having the change made in pg_index would mean physical replicas have the index disabled too. For the primary use case I have in mind (test disabling indexes you're considering dropping), having the disabledness replicate would be very useful. David
